Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mobility Accelerates Learning: Convergence Analysis on Hierarchical Federated Learning in Vehicular Networks

Tan Chen, Jintao Yan, Yuxuan Sun, Sheng Zhou, Deniz Gündüz, Zhisheng Niu

TL;DR

This work analyzes hierarchical federated learning in vehicular networks under mobility, deriving convergence bounds that reveal mobility can accelerate learning by fusing edge data and shuffling edge models. For classification tasks, mobility-specific mobility factors are derived, and ring-topology analysis yields closed-form eigenvalues linking speed and convergence. Empirical results on CIFAR-10 show mobility can boost accuracy by up to 15.1% in edge-non-i.i.d. settings, with faster speeds generally improving convergence up to a saturation point. The findings underscore mobility as a resource that, when harnessed appropriately, reduces training time and improves model performance in privacy-preserving, edge-assisted learning for connected vehicles.

Abstract

Hierarchical federated learning (HFL) enables distributed training of models across multiple devices with the help of several edge servers and a cloud edge server in a privacy-preserving manner. In this paper, we consider HFL with highly mobile devices, mainly targeting at vehicular networks. Through convergence analysis, we show that mobility influences the convergence speed by both fusing the edge data and shuffling the edge models. While mobility is usually considered as a challenge from the perspective of communication, we prove that it increases the convergence speed of HFL with edge-level heterogeneous data, since more diverse data can be incorporated. Furthermore, we demonstrate that a higher speed leads to faster convergence, since it accelerates the fusion of data. Simulation results show that mobility increases the model accuracy of HFL by up to 15.1% when training a convolutional neural network on the CIFAR-10 dataset.

Mobility Accelerates Learning: Convergence Analysis on Hierarchical Federated Learning in Vehicular Networks

TL;DR

This work analyzes hierarchical federated learning in vehicular networks under mobility, deriving convergence bounds that reveal mobility can accelerate learning by fusing edge data and shuffling edge models. For classification tasks, mobility-specific mobility factors are derived, and ring-topology analysis yields closed-form eigenvalues linking speed and convergence. Empirical results on CIFAR-10 show mobility can boost accuracy by up to 15.1% in edge-non-i.i.d. settings, with faster speeds generally improving convergence up to a saturation point. The findings underscore mobility as a resource that, when harnessed appropriately, reduces training time and improves model performance in privacy-preserving, edge-assisted learning for connected vehicles.

Abstract

Hierarchical federated learning (HFL) enables distributed training of models across multiple devices with the help of several edge servers and a cloud edge server in a privacy-preserving manner. In this paper, we consider HFL with highly mobile devices, mainly targeting at vehicular networks. Through convergence analysis, we show that mobility influences the convergence speed by both fusing the edge data and shuffling the edge models. While mobility is usually considered as a challenge from the perspective of communication, we prove that it increases the convergence speed of HFL with edge-level heterogeneous data, since more diverse data can be incorporated. Furthermore, we demonstrate that a higher speed leads to faster convergence, since it accelerates the fusion of data. Simulation results show that mobility increases the model accuracy of HFL by up to 15.1% when training a convolutional neural network on the CIFAR-10 dataset.
Paper Structure (25 sections, 16 theorems, 67 equations, 9 figures, 1 table, 1 algorithm)

This paper contains 25 sections, 16 theorems, 67 equations, 9 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. HFL in Vehicular Networks
  4. Mobility-aware Hierarchical Federated Averaging
  5. Data Distribution
  6. Mobility Model
  7. Convergence Analysis
  8. Convergence of HFL in static and mobile scenarios
  9. Static Scenario
  10. Mobility Scenario
  11. Convergence of HFL for classification tasks
  12. Mobility Factor for Classification Problem
  13. Mobility Factor for Ring Topology
  14. Simulation Results
  15. Simulation Settings
  16. ...and 10 more sections

Key Result

Proposition 1

For any $m$, if assumption assu1 holds, and for some $\epsilon\ge 0$, we have after $T\triangleq K\tau_l\tau_e$ local updates, the loss function of HFL is bounded by

Figures (9)

  • Figure 1: Illustration of HFL in vehicular networks.
  • Figure 2: The training procedure of HFL.
  • Figure 3: Simulation scenario: HFL with one cloud server, $N=4$ edge servers, and $M=32$ vehicles. The edge servers form a ring topology to compose a square.
  • Figure 4: Average sojourn probability of vehicles with different vehicle speeds. The length of each side of the road is set to $a=1000$ m.
  • Figure 5: Comparison of the test accuracy within 600 cloud epochs for different initial data distribution.
  • ...and 4 more figures

Theorems & Definitions (30)

  • Proposition 1
  • proof
  • Theorem 1
  • proof
  • Theorem 2
  • proof
  • Definition 1
  • Proposition 2
  • proof
  • Lemma 1
  • ...and 20 more